Earthq Sci (2011)24: 77–85 77

doi:10.1007/s11589-011-0771-9

Spatial variations of the 660-km discontinuityin the western Pacific subduction zones

observed from CEArray triplication data∗

Baoshan Wang1, and Fenglin Niu2

1 Institute of Geophysics, China Earthquake Administration, Beijing 100081, China2 Department of Earth Science, Rice University, Houston, TX 77005, USA

Abstract We examined the spatial variation of velocity structures around the 660-km discontinuity at the

western Pacific subduction zones by waveform modeling of triplication data. Data from two deep earthquakes

beneath Izu-Bonin and Northeast China are used. Both events were well recorded by a dense broadband seismic

network in China (CEArray). The two events are located at approximately the same distance to the CEArray, yet

significant differences are observed in their records: (1) the direct arrivals traveling above the 660-km discontinuity

(AB branch) are seen in a different distance extent: ∼29◦ for the NE China event, ∼23◦ for Izu-Bonin event;

(2) the direct (AB) and the refracted waves at the 660-km (CD branch) cross over at 19.5◦ and 17◦ for the NE

China and the Izu-Bonin event, respectively. The best fitting model for the NE China event has a broad 660-km

discontinuity and a constant high velocity layer upon it; while the Izu-Bonin model differs from the standard

IASP91 model only with a high velocity layer above the 660-km discontinuity. Variations in velocity models can

be roughly explained by subduction geometry.

Key words: P-wave triplication; 660-km discontinuity; western Pacific subduction zone

CLC number: P315.2 Document code: A

1 Introduction

The 660-km seismic discontinuity dividing the

upper and lower mantle plays a significant role in large-

scale mantle circulation (e.g., Tackley, 2008). The seis-

mic discontinuity is generally believed to be caused

by the dissolution of γ-olivine or ringwoodite into per-

ovskite plus magnesiowustite (Ito and Takahashi, 1989).

In general, the phase transition (also known as the post-

spinel transformation) is observed to occur within a rel-

atively narrow pressure range, consistent with a sharp

660-km discontinuity observed seismically. Laboratory

experiments also showed that ringwoodite tends to

dissolve under higher-pressure and lower-temperature

conditions, consistent with depressed 660-km disconti-

nuities seen in cold subduction regions (e.g., Shearer

∗ Received 20 September 2010; accepted in revised form 22

December 2010; published 10 February 2011.

Corresponding author. e-mail: wangbs@cea-igp.ac.cn

The Seismological Society of China and Springer-Verlag Berlin

Heidelberg 2011

and Masters, 1992; Vidale and Benz, 1992; Niu et al.,

2005).

Several receiver function studies (e.g., Niu and

Kawakatsu, 1996; Ai et al., 2003; Andrews and Deuss,

2008) indicated that the 660-km discontinuity beneath

subduction regions has complicated structure featured

by multiple velocity jumps rather than a simply de-

pressed discontinuity. Recently, Wang and Niu (2010)

found a broad 600-km discontinuity (velocity increases

gradually in a depth range of 50–70 km) beneath NE

China with waveform triplication data recorded by a

dense broadband seismic network in China. The broad-

ened discontinuity was interpreted to be caused by

multiple phase transitions associated with the disso-

lutions of olivine and garnet within a cold subducted

slab. Numerical calculation of mantle velocity (Vacher

et al., 1998) also indicated the multiple phase changes

could result in a series of velocity jumps in the depth

range of ∼660–720 km. The geodynamic implication of

such a broad 660-km is, however, not well understood.

It is also unclear how pervasive such a broadened 660-

78 Earthq Sci (2011)24: 77–85

km discontinuity is present at various sections of the

subducting Pacific slab.

The Pacific Plate have been subducting beneath

Japan and the Izu-Bonin arc in the last 50 Ma. At

the northern section, the Pacific Plate is subducting

beneath the Eurasian Plate at a speed ∼10 mm/a with

a relatively low descending angle. At the southern sec-

tion, the Pacific Plate is descending at a near vertical

angle beneath the Philippine Sea Plate (e.g., Fukao et

al., 2001; Nakajima and Hasegawa, 2007). Moderate

to large deep focus earthquakes occur within both sec-

tions of the subducting Pacific Plate. As the region

is relatively well instrumented, it is thus an idea place

to study the fine seismic structure at the upper-lower

mantle boundary (e.g., Niu and Kawakatsu, 1996; Ai et

al., 2003; Tajima et al., 2009; Wang and Chen, 2009).

One of the most frequently used methods in study-

ing the 660-km discontinuity is forward modeling of the

wave bifurcation arising from the discontinuity recorded

at epicentral distances between ∼10◦ and 30◦. The mul-

tiple arrivals recorded at a station share very similar ray

paths near the source and the receiver sides, thus the

differential travel time and relative amplitude among

these arrivals are not sensitive to shallow structures.

This method has been widely used to study the 660-km

discontinuity beneath various subduction zones (e.g.,

Tajima and Grand, 1998; Brudzinski and Chen, 2003;

Wang et al., 2006; Wang and Chen, 2009).

To better monitor seismic activities in China,

the China Earthquake Administration (CEA) has up-

graded and expanded its national and regional networks

over the last decade. Currently, CEA owns one of the

largest permanent broadband seismic network in the

world, which consists of one national network, 31 re-

gional networks, and several other networks. The net-

works can be integrated as a large seismic array, which

will be referred to as CEArray in the remainder of

this paper. CEArray include more than 1 000 stations

(Figure 1), among which ∼900 are broadband (Zheng

et al., 2009). In a recent study, we (Wang and Niu,

2010, hereafter referred to as WN10) analyzed CEArray

data from a deep earthquake occurred near the border

among China, Russia and North Korea to investigate

lateral variations around the 660-km beneath Northeast

China. Here we used a similar approach to study the

triplication data from a deep earthquake occurring at

the Izu-Bonin region. We then compared the velocity

structure obtained here with those determined by

WN10 to study lateral variations of the 660-km dis-

continuity.

South ChinaSea Islands

Izu-Bonin Eq.

NE China Eq.

E

N

Figure 1 Map showing the stations of the national and regional networks operated by the China Earthquake

Administration (solid triangles). Also shown is the focal mechanism of two deep earthquakes used in this study

(beach balls). Contours of upper boundary of the subducting Pacific slab are outlined by dashed gray lines.

Earthq Sci (2011)24: 77–85 79

2 Data and method

A sudden increase of seismic velocity within man-

tle could result in multiple arrivals of P- and S-waves

at a single distance, leading to the so-called waveform

triplication. The wave chain includes the direct arrival

traveling above the seismic discontinuity, the waves re-

flected and refracted at the seismic discontinuity. In

Figure 2, we showed an example of the ray paths of

the three arrivals associated with the 660-km discon-

tinuity, together with their arrival times from a deep

earthquake. As shown in Figure 2a, ray paths for di-

rect (AB), reflected (BC), and refracted (CD) waves are

quite close to each other at the uppermost mantle and

crust. Thus the relative moveouts and amplitude con-

trast between the direct and refracted phases are very

sensitive to velocity structure right above and below the

x/°

(a)

(b) A

B

O

C

D

surface

km

km

Direct (AB) Refracted (CD)

Reflected (BC)

t x s

Station

Source

Figure 2 Schematic ray paths of the multiple arrivals

(a) and the calculated travel times of these arrivals (b).

x denotes the distance, which also have the same mean-

ing in Figures 4, 6, 7. A 520 km deep source and the

IASP91 model were used in calculating the travel times.

The direct AB phase, the reflected BC phase, and the re-

fracted CD phase are shown in solid, dotted, and dashed

lines, respectively.

discontinuity (e.g., Tajima and Grand, 1998; Brudzinski

and Chen, 2003; Wang et al., 2006; Wang and Chen,

2009; Wang and Niu, 2010).

We found a total of eight earthquakes occurring

between 2008 and 2009 with a focal depth >410 km

and magnitude >5 within the triplication distances of

the CEArray. To analyze the triplication induced by

the 660-km discontinuity, we used following criteria to

select the events. Events with moderate magnitude are

preferred, because they are strong enough to be clearly

recorded by most stations and also they usually have

simple source time functions. On the other hand, to

fully track the triplication we required a full coverage of

recordings in the distance range between 10◦ and 30◦.Based on these requirements, we selected two earth-

quakes for further study.

The first event is located near the border of east

Russia and Northeast China, hereafter referred to as

NE China event (Figure 1). The NE China event oc-

curred on May 19, 2008 with a focal depth of 520 km

and a moment magnitude 5.7 (Harvard CMT solution).

Waveforms from NE China event were modeled by

WN10, and here for completeness we reiterate the main

features of the triplication observed from this event:

(1) the AB branch extends to as far as ∼29◦; (2) the

CD phase does not show up until ∼14◦–15◦ away from

the earthquake; (3) the crossover of the AB and CD

branches occurs between 19◦ and 20◦.The second earthquake occurred on July 20, 2008

in the Izu-Bonin arc (Figure 1) with a focal depth of

481 km (USGS PDE catalog) and a moment magnitude

of 5.8 (Harvard CMT solution), hereafter we refer to it

as the Izu-Bonin event. Seismic waves generated by the

Izu-Bonin event were clearly registered by more than

500 broadband stations of the CEArray. The CEArray

stations are in the azimuth range of ∼260◦–350◦. Basedon the waveform coherence, we divided the CEArray

stations into two azimuthal groups, one in the azimuth

range of 260◦–297◦ and the other in 326◦–350◦. This

division is consistent with the focal mechanism of the

event, which shows a 313◦ strike for one of the fault

planes (Figure 3). Stations with an azimuth close to

313◦ were not selected so as to avoid modeling compli-

cated waveforms.

Before fitting the waveform data, we first removed

instrumental responses from the data and further fil-

tered the waveform data with a bandpass filter (0.1–1.0

Hz). To represent the data in a comprehensible way,

we plotted recording section by aligning the manually

picked first arrivals to predictions from certain velocity

80 Earthq Sci (2011)24: 77–85

E

N

W

S

T

X

P

Mrr Mtt

Mff

Mrt Mrf

Mtf

Figure 3 Distribution of normalized amplitudes for the

P wave radiated from the Izu-Bonin event are shown as

function of takeoff angle and azimuth. “P” and “T” are

the compressional and tensional axis. Number 001–006

indicate the azimuths 240◦–340◦.

model, as done by most 1D forward modeling (e.g., Taji-

ma and Grand, 1998; Wang and Chen, 2009; Wang and

Niu, 2010). Figure 4 shows examples of waveforms from

the two azimuthal groups. In general, the first recording

section (Az 260◦–297◦) has lower SNR (signal-to-noise

ratio) than the second one. We also noticed there is a

significant difference in waveform durations. While the

waveforms in the second group (Az 326◦–350◦) have a

half duration of ∼2 s, close to the estimate of its CMT

solution; those from the first section (Az 260◦–297◦) aremuch broader, with a half duration of ∼4 s. A broad-

ened P wave was also observed by Tajima and Nakagawa

(2006). Waveforms in the first section also appeared to

be less coherent than those in the second group. This

is probably due to strong lateral heterogeneity along

the ray paths of these low azimuth stations. Since 3D

structure may play an important role in generating the

complicated waveforms in this group, it may not be ap-

propriate to use 1D or 2D modeling techniques to fit

the waveform data. We thus decided to only model the

waveform data of the second group (Figure 4b).

Clear triplication can be seen in the recording sec-

tion and can be summarized as following: (1) the AB

phase extends no longer than 23◦; (2) the CD has am-

plitude comparable to the AB from the beginning of the

recording section (∼16.5◦); (3) the AB and CD braches

crossover at around 17.5◦. The triplication features

shown here is very different from what we observed from

the NE China event, they are rather close the global av-

erage model, IASP91 (Kennett and Engdahl, 1991).

(a) Az ° °

x°

t x s

x°

(b) Az ° °

t x s

Figure 4 Record sections for the Izu-Bonin event at two different azimuth ranges. For display purpose, only

a small portion of the seismograms is shown here. The ray-theory based travel times predicted from IASP91 are

plotted in dashed lines. Note here we adjusted the first arrival times of the data in order to align the observed

first arrivals with the model predictions. The adjustments are very minor, usually less than 0.7 s.

Earthq Sci (2011)24: 77–85 81

3 Waveform modeling

In this paper, we follow the analyzing strategy

and procedure of WN10. Since uncertainty in earth-

quake location especially in focal depth may affect the

modeling result. We redetermined the focal depth by

using the global records of this earthquake (with epi-

central distance 30◦–90◦) collected by the Incorporated

Research Institute for Seismology (IRIS). We manually

picked the P and pP arrival times from the records

and used a grid search method to find the focal depth

that minimizes the misfit between observed and IASP91

predicted pP–P differential travel time. As shown in

Figure 5, the misfit reaches a minimum at the depth of

481 km, identical to the PDE catalog depth.

Based on the analysis of the NE China event, we

found that the maximum reachable distance of the di-

rect AB phase is sensitive to the velocity gradient right

above the discontinuity. The minimum distance of the

CD phase is closely related to the sharpness and amount

of velocity jumps across the discontinuity. Based on the

first feature that the extension of the AB phase is com-

parable to the IASP91 model (Figure 4b), we started

with an initial model that has a velocity gradient sim-

ilar to the IASP91 model for layers above the 660-km.

In fitting the data, we did not intend to match the

arrival time of the first arrival, but rather emphasized

fitting the relative move-out and amplitude of the AB

and CD phases. After several iteration of trial and

error fitting, we obtained a final 1D velocity that fits

the observed waveform data reasonably well (Figure

6). We understand that there are many models that

can explain the data within the data uncertainty, and

we have chosen a relatively simple model here as the

final model. We computed the synthetic seismograms

with a modified propagation matrix method proposed

by Wang (1999). Synthetics from two rather simple

1D velocity models explain the observed features of the

two earthquakes quite well. The time shift between the

manually picked and the synthetic first arrivals are usu-

ally less than 0.7 s. Some minor misfits can be found at

epicentral distances less than 17◦ for Izu-Bonin event

pPP

time

resi

dual

/s

Depth/km

Figure 5 pP–P differential time residuals with respect

to the IASP91 model are shown as a function of assumed

hypocentral depth.

Dep

th/k

m

P velocity/(km.s )

Izu-BoninNE ChinaIASPDepressed

(a) Izu-Bonin (b) NE China (c) vP model

x x

t x s t x s

Figure 6 Synthetic waveforms (gray) computed from velocity models for Izu-Bonin event (a) and NE China event (b).

(c) Final velocity models for both events are shown together with IASP91 model, also shown is a velocity model with

a depressed 660-km discontinuity to 670 km. Observations were aligned with synthetics at the first arrivals. Note the

excellent match between the synthetics and the observed waveforms.

82 Earthq Sci (2011)24: 77–85

and ∼23◦ for NE China event (Figure 6). So the true

models of the two regions must have finer structures

than the two relatively simple 1D models shown in Fig-

ure 6c. In addition, velocity of the layer immediately

below the earthquakes is not well constrained due to

limits either in data coverage or in the method. On

the other hand, we have very good coverage around the

crossover distance of the two phases, so velocity in the

layers right above or below the discontinuity is relatively

well constrained.

4 Results and discussion

The final velocity models from the those two

events show distinct differences (Figure 6c). For the

velocity right above the 660-km, the NE China event

revealed a layer with an almost constant velocity, i.e.

low gradient, while from the Izu-Bonin event we mod-

eled a layer with velocity gradient similar to the IASP91

model. As mentioned above, the gradient is determined

from the maximum epicentral distance that the AB

phase is able to reach, which is ∼23◦ and ∼29◦ for Izu-

Bonin event and NE China event, respectively. Records

from the NE China event is well explained by a broad

(50–70 km) 660-km discontinuity. Models with a ve-

locity jump at the depth of 660 km seem to be able

to fit the waveform data of the Izu-Bonin event. On

the other hand, velocity increase across the 660-km dis-

continuity is determined by the differential travel times

and the relative amplitudes of the AB and CD phases.

As shown in Figure 6, the differential travel times and

relative amplitudes are well matched for both events.

Spatial variations of the 660-km discontinuities

across various sections of the subduction zones in the

western Pacific were studied with a few stations sparsely

distributed in China (e.g., Wang et al., 2006; Tajima

et al., 2009). As shown in the records of the Izu-Bonin

event, waveforms recorded at different azimuths could

be very different due to either a difference in radiation

pattern or strong lateral heterogeneities. Thus it may

be not appropriate to combine records with different

azimuths into a single record and model them with a

1D velocity model. With the availability of the CEAr-

ray data, we were able to group data within a relatively

small azimuthal range. Such grouping can minimize un-

certainties induced by unmodeled 3D velocity structure

and is obviously more favorable for 1D waveform mod-

eling. With this grouping, we were able to image subtle

lateral variations from the NE China event (Wang and

Niu, 2010).

A high velocity at transition zone depth is usually

interpreted to have a thermal origin associated with

subducting slabs. The cold slabs when approaching

to the bottom of the upper mantle, can also delay the

post-spinel phase change, leading to a depressed 660-km

(e.g., Wang et al., 2006; Tajima et al., 2009; Wang and

Chen, 2009). From the Izu-Bonin event, we obtained a

model with a slightly high velocity above the 660-km

x°

t x s

(a) Depressed

x°

t x s

(b) IASP91

Figure 7 Comparison between observed and synthetic waveforms from the model with a depressed 660-km

discontinuity (a) and IASP91 standard model (b).

Earthq Sci (2011)24: 77–85 83

which shows no depression. We believe both the high ve-

locity and the discontinuity depth are well constrained

from the data. For comparison, we calculated synthet-

ics of a 1D model modified from the final Izu-Bonin

model. The model has a depressed 660-km at 670 km

depth (Figure 6c). Synthetic waveforms corresponding

to this model are shown in Figure 7, together with the

IASP91 synthetics. The synthetics from the depressed

660-km model showed a poor fit with records before the

crossover distance (Figure 7a). On the other hand, the

IASP91 synthetics had large misfits with waveforms

recorded at large distances (Figure 7b). Tajima et

al. (2009) also proposed a similar velocity model (M2.0)

with a 660-km discontinuity at its normal depth and

Figure 8 Ray paths for AB (a) and CD (b) branches are plotted on top of tomographic images of Fukao et al. (2001).

Ray paths for Izu-Bonin event and NE China event are shown in red and green, respectively. Dots indicate the

earthquakes and the dashed gray lines represent the Wadati-Benioff zone of the subducting Pacific slab.

a high velocity above the discontinuity for this region.

The ray paths of the AB and CD arrivals at two

depth ranges are shown on top of the tomographic im-

age from Fukao et al. (2001) (Figure 8). Systematic

differences can be found for both phases, ray paths from

the NE China event fall mainly inside the high velocity

regions, while seismic waves from the Izu-Bonin event

travels along the northeastern edge of a high velocity

body. To further understand the difference shown in

the two final 1D models, in Figure 9 we showed two

velocity sections along the AA′ and BB′ lines in Figure

8. We also showed the ray paths of the direct, reflected

and refracted waves recorded at 20◦ (Figure 9). High

velocity anomalies beneath Northeast China (Figure 8

and Figure 9b) were interpreted as the subducted Pa-

cific Plate lying above the 660-km discontinuity (Fukao

et al., 2001). This high velocity anomaly is well sampled

by the AB phase. On the contrary, for the Izu-Bonin

event both AB and CD phases only touch the lower

boundary of the Pacific Plate (Figure 9a). Little to

no anomaly was seen at the turning depths of the two

phases, which may explain the normal behavior of the

660-km in this region.

The velocity distributions shown in Figure 9 may

be a result of subduction slab geometry. The Pacific

Figure 9 Vertical cross section along AA′ and BB′ inFigure 8, ray paths for AB, BC, and CD calculated from

the corresponding final models (Figure 6c) are shown in

solid lines.

84 Earthq Sci (2011)24: 77–85

Plate is subducting at a high angle beneath Izu-Bonin

region. The stations used for Izu-Bonin event are lo-

cated at the direction almost normal to the westward

subducting, and seismic waves recorded by those sta-

tions sample the subduction slab before its deflection at

the bottom of the upper mantle (Figure 8). In contrast,

seismic waves from the NE China event sample mainly

the flattened part of the slab. As a result the high ve-

locity anomaly shown in Izu-Bonin region is ∼100 km

shallower than that beneath NE China (Figure 9). Ac-

cording to the subduction geometry, we would expect a

different structure at lower azimuth ranges in Izu-Bonin

region (Tajima et al., 2009) where the Pacific slab may

start to lie atop the 660-km discontinuity.

Many studies revealed a depressed 660-km beneath

NE China from triplication waveform modeling (e.g.,

Wang et al., 2006; Tajima et al., 2009; Wang and Chen,

2009). On the other hand, we found that depression

is probably the low-resolution image of broadened dis-

continuity at depth range between 660 km and 720

km. Such a gradual change in velocity is likely caused

by a series phase transitions of mantle minerals under

low temperature condition (Vacher et al., 1998). We

found large variations in the waveform data from the

Izu-Bonin earthquake, indicating strong lateral het-

erogeneities are present near the earthquake source

and along the ray paths. Modeling waveform records

with high azimuthal angles and raypaths sampling the

lower boundary of the Pacific slab results in a relatively

“normal” velocity structure located northeast to the

Izu-Bonin arc.

5 Conclusions

We investigated spatial variations of velocity

structure around the 660-km discontinuity in west-

ern Pacific subduction based on used triplication data

recorded by CEArray from two deep focus earthquakes.

The high station density of the CEArray allowed us to

use stations within a narrow azimuth to better constrain

1D velocity structures. We found a very broad 660-km

beneath NE China and relatively normal mantle below

the Pacific Plate in the northeast flank of the Izu-Bonin

arc. The difference in the P-wave velocity structure

seems to be caused mainly by the temperature anoma-

lies posted by the subducting Pacific Plate. Further

studies with more events and on the S-wave velocity

structure around the 660-km discontinuity will certainly

help to clarify this interpretation.

Acknowledgements Waveform data for this

study were provided by Data Management Centre of

the China National Seismic Network at Institute of Geo-

physics, China Earthquake Administration. We thank

Prof. Rongjiang Wang for making his code available to

us. Critical comments from two anonymous reviewers

significantly improved the quality of this paper. This

work was supported by National Natural Science Foun-

dation of China under grant 40874095 and NSF under

grant EAR-063566.

References

Ai Y, Zheng T, Xu W, He Y and Dong D (2003). A complex

660 km discontinuity beneath northeast China. Earth

Planet Sci Lett 212: 63–71.

Andrews J and Deuss A (2008). Detailed nature of the

660 km region of the mantle from global receiver func-

tion data. J Geophys Res 113: B06304, doi:10.1029/

2007JB005111.

Brudzinski M R and Chen W P (2003). A petrologic

anomaly accompanying outboard earthquakes beneath

Fiji-Tonga: Corresponding evidence from broadband

P and S waveforms. J Geophys Res 108, doi:

1029/2002JB002012.

Fukao Y, Widiyantoro S and Obayashi M (2001). Stagnant

slabs in the upper and lower mantle transition region.

Rev Geophys 39: 291–323.

Ito E and Takahashi E (1989). Postspinel transformations

in the system Mg2SiO4-Fe2SiO4 and some geophysical

implications. J Geophys Res 94: 10 637–10 646.

Kennett B L N and Engdahl E R (1991). Traveltimes for

global earthquake location and phase identification.

Geophys J Int 105: 429–465, doi:10.1111/j.1365-246X.

1991.tb06724.x.

Nakajima J and Hasegawa A (2007). Subduction of the

Philippine Sea plate beneath southwestern Japan: Slab

geometry and its relationship to arc magmatism. J Geo-

phys Res 112: B08306, doi:10.1029/2006JB004770.

Niu F and Kawakatsu H (1996). Complex structure of man-

tle discontinuities at the tip of the subducting slab be-

neath Northeast China—A preliminary investigation of

broadband receiver functions. J Phys Earth 44: 701–

711.

Niu F, Levander A, Ham S and Obayashi M (2005). Map-

ping the subducting Pacific slab beneath southwest

Japan with Hi-net receiver functions. Earth Planet Sci

Lett 239: 9–17.

Shearer P M and Masters T G (1992). Global mapping of

topography on the 660-km discontinuity. Nature 355:

791–796.

Tackley P J (2008). Layer cake or plum pudding? Nature

Geoscience 1: 157–158.

Earthq Sci (2011)24: 77–85 85

Tajima F and Grand S P (1998). Variation of transition

zone high-velocity anomalies and depression of 660 km

discontinuity associated with subduction zones from the

southern Kuriles to Izu-Bonin and Ryukyu. J Geophys

Res 103: 15 015–15 036.

Tajima F and Nakagawa T (2006). Implications of seis-

mic waveforms: Complex physical properties associated

with stagnant slab. Geophys Res Lett 33: L03311, doi:

10.1029/2005GL024314.

Tajima F, Katayama I and Nakagawa T (2009). Variable

seismic structure near the 660 km discontinuity associ-

ated with stagnant slabs and geochemical implications.

Phys Earth Planet Inter 172: 183–198.

Vacher P, Mocquet A and Sotin C (1998). Computation of

seismic profiles from mineral physics: the importance of

the non-olivine components for explaining the 660 km

depth discontinuity. Phys Earth Planet Inter 106: 275–

298.

Vidale J and Benz H (1992). Upper-mantle seismic discon-

tinuities and the thermal structure of subduction zones.

Nature 356: 678–680.

Wang B and Niu F (2010). A broad 660-km discontinu-

ity beneath northeast China revealed by dense regional

seismic networks in China. J Geophys Res 115: B06308,

doi:10.1029/2009JB006608.

Wang R (1999). A simple orthonormalization method for

stable and efficient computation of Green’s functions.

Bull Seismol Soc Am 89: 733–741.

Wang T and Chen L (2009). Distinct velocity variations

around the base of the upper mantle beneath northeast

Asia. Phys Earth Planet Inter 172: 241–256.

Wang Y, Wen L, Weidner D and He Y (2006). SH velocity

and compositional models near the 660-km discontinuity

beneath South America and northeast Asia. J Geophys

Res 111: B07305, doi:10.1029/2005JB03849.

Zheng X F, Ouyang B, Zhang D N, Yao Z X, Liang J H

and Zheng J (2009). Technical system construction of

Data Backup Centre for China Seismograph Network

and the data support to researches on the Wenchuan

earthquake. Chinese J Geophys 52: 1 412–1 417, doi:

10.3969/j.issn.0001-5733 (in Chinese).